Stay connected with KayaToday—follow us on Instagram and Facebook for the latest news and reviews delivered straight to you.
There is a kind of capital allocation decision that reveals more about a company’s true constraints than any earnings call. When the world’s most digitally native companies — built on software, cloud computing, and increasingly artificial intelligence — begin signing multi-decade contracts with nuclear power plants, the decision is not a sustainability gesture. It is a confession about supply.
On June 11, Amazon Web Services announced a power purchase agreement with Talen Energy for 1.9 gigawatts of electricity, running through 2042, drawn from the Susquehanna nuclear plant in Pennsylvania. The arrangement builds on an existing relationship and supports Amazon’s plan to invest $20 billion in AWS facilities in the state. It is the third major nuclear deal announced by a hyperscale technology company since early May, following commitments from Google and Meta. None of these companies built their businesses around energy infrastructure. All of them are now behaving like utilities.
The Demand Curve That Broke the Old Plan
The proximate cause is not complicated. Global data centre electricity consumption is projected to grow from approximately 460 terawatt-hours in 2024 to 1,300 terawatt-hours by 2035 — nearly a threefold increase within a single decade, driven almost entirely by the computational demands of training and running AI models. That is not a curve that conventional grid expansion, regulatory approval timelines, and power plant construction schedules were designed to accommodate. Solar and wind, for all the genuine progress in deployment cost and speed, are intermittent by nature — a poor match for data centres that require continuous, predictable power twenty-four hours a day, every day of the year, indefinitely.
Read also this: https://kayatoday.com/news/jensen-huang-did-not-come-to-taipei-to-sell-chips-he-came-to-declare-an-empire/
Nuclear power offers what the AI industry’s power appetite specifically requires: high energy density, minimal land footprint relative to output, and — critically — a baseload generation profile that does not fluctuate with weather. The arithmetic that has pushed Amazon, Google, Meta, and Microsoft toward nuclear power is not ideological. It is operational necessity dressed in the language of decarbonisation.
Two Strategies, One Constraint
What is emerging across the industry is a bifurcated approach: restart what already exists, and invest in what does not yet exist commercially at scale.
Microsoft’s 20-year power purchase agreement with Constellation Energy to restart Three Mile Island’s Unit 1 — the reactor that was not involved in the 1979 accident, shut down in 2019 for economic reasons rather than safety ones — exemplifies the first strategy. A dormant, fully licensed nuclear asset, sitting idle for want of a buyer willing to underwrite its economics, suddenly became commercially viable the moment a technology company offered the kind of long-term revenue certainty that justified the restart investment. Constellation has been moving through the regulatory steps to bring the facility back online throughout 2025 and into this year.
The second strategy — investment in next-generation small modular reactors, or SMRs — is the more genuinely novel bet. Google’s agreement with Kairos Power targets up to 500 megawatts of SMR capacity beginning around 2030. Amazon has backed 5 gigawatts of X-energy SMR projects and led a $500 million financing round to help the company finish its reactor design and build a nuclear fuel fabrication facility. Oracle’s Larry Ellison announced plans for a gigawatt-scale data centre powered by three SMRs, claiming the necessary building permits were already secured. Meta has issued a request for proposals targeting 1 to 4 gigawatts of new nuclear generation, open to both SMRs and larger conventional reactors.
SMRs are not yet a proven commercial technology at the scale these companies are betting on. Their appeal lies in modular factory production, smaller land requirements, and theoretically faster deployment timelines than traditional gigawatt-scale reactors — but none of the major SMR designs backing this wave of corporate investment has yet been deployed commercially in the United States. The earliest projected operational dates cluster around 2030, meaning the technology companies underwriting these projects are making capital commitments years ahead of any proof that the underlying engineering can be delivered on schedule and on budget — a pattern that has bedevilled nuclear construction projects of every generation.
What This Means Beyond Silicon Valley
The strategic significance of this shift extends well beyond the balance sheets of four technology companies. By providing the revenue certainty that nuclear projects have historically struggled to secure from conventional utility customers, hyperscalers are becoming the primary financiers of a nuclear renaissance that public policy alone had failed to deliver for two decades. That is a remarkable inversion of the traditional relationship between technology companies and energy infrastructure — and it concentrates an unusual degree of influence over the pace and location of nuclear development in the hands of a small number of private corporations whose primary motivation is securing power for AI workloads, not advancing a national energy strategy.
For countries and regions competing to host the next generation of AI infrastructure investment — including the connector economies of Southeast Asia that have featured prominently in the broader reconfiguration of global supply chains — the lesson is direct. Data centre investment is no longer simply a matter of land, labour, and connectivity. It is, increasingly, a matter of which jurisdictions can offer credible, long-term access to firm power at the gigawatt scale. Few currently can. Fewer still have a regulatory pathway to nuclear power that could plausibly compete with what Pennsylvania, Virginia, and Washington State are now offering. That gap, more than any tax incentive, may end up determining where the next wave of AI infrastructure actually gets built.
Read Also: The Emerging Technologies That Will Outlast the AI Moment
